Start here
A map, organized by the problem you actually have
Measurement problems rarely arrive labeled with their method. They arrive as two dashboards disagreeing, or a campaign that tested well and did nothing. This page routes from the symptom to the part of the curriculum that covers it.
The one idea that organizes everything else
Every marketing measurement question is a question about a comparison. Not "how many conversions did this campaign get," but "how many more than what would have happened otherwise."
That second thing, the counterfactual, is never observed. It did not happen. Every method on this site is a different strategy for building a credible stand-in for it, and every method's weaknesses are the weaknesses of its stand-in.
Attribution does not build one at all. That is a category difference rather than a flaw. It answers a useful question, which touchpoints preceded the conversions we can see, and that question is not the one about incremental value.
The progression
Six rungs. Most programs are strong on the first two, weak on the third, and jump straight to the fifth.
01
Observe
What happened?
02
Describe
Where did conversions appear to come from?
03
Estimate
What changed because of marketing?
04
Explain
How did channels and outside factors contribute?
05
Decide
What should happen next?
06
Learn
What uncertainty should we reduce next?
Where should you start?
Find the symptom that sounds most like your week.
-
“Two systems report different numbers for the same campaign.”
Almost always a definition or attribution-window difference rather than a bug. Platforms count conversions they can claim, in windows they choose, against identities they can resolve. Reconcile the definitions before the numbers.
Start with attribution. → -
“Our ROAS looks great but revenue is flat.”
The signature of measuring credit rather than causation. Reported ROAS can rise while incremental revenue falls, because the channels that harvest existing demand are the ones best positioned to claim it.
Read Attribution Is Not Incrementality. → -
“We want to know if a channel is worth the spend.”
This is an incrementality question, so it needs a holdout. The design work matters more than the estimator: what you can withhold, from whom, and for how long.
Work through the geo holdout example. → -
“Our model has great AUC but the campaign did not work.”
You targeted propensity when you needed persuadability. A model that finds who is likely to convert will spend your budget on people who were converting anyway.
Head to decisioning. → -
“Leadership wants one number for marketing effectiveness.”
There is not one. There is a set of estimates with different assumptions and different uncertainty. The useful move is triangulation: take the agreement where methods agree, and be specific about where they do not.
See how the evidence types differ. →
Or start with a discipline
Attribution
Can we trust what we observe?
Tracking loss, identity, consent, event definitions, funnel grain, and attribution as descriptive evidence. Everything downstream inherits the flaws you accept here.
Incrementality
Did marketing change the outcome?
Counterfactuals, experiments, quasi-experiments, and the diagnostics that decide whether an estimate deserves to be believed. Causal ML sits here too, where the question moves from an average effect to whose effect differs.
Allocation
Where should the next dollar go?
Marginal rather than average returns, channel interaction, marketing mix modeling, and budget decisions made under real uncertainty.
Decisioning
What should happen after the click?
The optimization signal ladder, lifecycle experimentation, uplift modeling, and the systems that turn estimates into governed decisions. Where causal ML stops being an estimate and becomes a policy.